Data-Driven Risk Predictions for COVID-19

Abstract

Effective medical decision making requires an understanding of patient risk. This is especially critical in the COVID-19 pandemic where resource scarcity necessitates patient prioritization and care management. In this work, we propose an international COVID-19 mortality risk (CMR) calculator for hospitalized patients. Through a multi-center study leveraging data from 33 hospitals and nearly 4000 patients across Europe and the United States, we train and validate a machine learning model that confirms previously identified clinical risk factors. The model also demonstrates strong quantitative performance (AUCs 0.81-0.92) on external populations. The tool is accessible to clinicians through an online application and is currently in use at a major hospital in Italy.

Date
Event
Session MD30: An Analytics Response to Covid-19
Location
INFORMS Annual Meeting 2020